Re: [PATCH 3/4] drm/amdgpu: update algorithm of umc address conversion

[Lazar, Lijo]

On 1/25/2022 4:16 PM, Tao Zhou wrote:

On ALDEBARAN, we need to traverse all column bits higher than
BIT11(C4C3C2) in a row, the shift of R14 bit should be also taken
into account. Retire all pages we find.

Signed-off-by: Tao Zhou <tao.zh...@amd.com>
---
  drivers/gpu/drm/amd/amdgpu/umc_v6_7.c | 41 +++++++++++++++++++++------
  drivers/gpu/drm/amd/amdgpu/umc_v6_7.h |  4 +++
  2 files changed, 37 insertions(+), 8 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/umc_v6_7.c 
b/drivers/gpu/drm/amd/amdgpu/umc_v6_7.c
index 300dee9ec6b4..1ecba7b5df1c 100644
--- a/drivers/gpu/drm/amd/amdgpu/umc_v6_7.c
+++ b/drivers/gpu/drm/amd/amdgpu/umc_v6_7.c
@@ -119,7 +119,7 @@ static void umc_v6_7_ecc_info_query_error_address(struct 
amdgpu_device *adev,
                                         uint32_t ch_inst,
                                         uint32_t umc_inst)
  {
-       uint64_t mc_umc_status, err_addr, retired_page;
+       uint64_t mc_umc_status, err_addr, soc_pa, retired_page, column;
        uint32_t channel_index;
        uint32_t eccinfo_table_idx;
        struct amdgpu_ras *ras = amdgpu_ras_get_context(adev);
@@ -145,15 +145,27 @@ static void umc_v6_7_ecc_info_query_error_address(struct 
amdgpu_device *adev,
                err_addr = REG_GET_FIELD(err_addr, MCA_UMC_UMC0_MCUMC_ADDRT0, 
ErrorAddr);

/* translate umc channel address to soc pa, 3 parts are included */

-               retired_page = ADDR_OF_8KB_BLOCK(err_addr) |
+               soc_pa = ADDR_OF_8KB_BLOCK(err_addr) |
                                ADDR_OF_256B_BLOCK(channel_index) |
                                OFFSET_IN_256B_BLOCK(err_addr);
+               /* clear [C4 C3 C2] in soc physical address */
+               soc_pa &= ~(0x7ULL << UMC_V6_7_PA_C2_BIT);

/* we only save ue error information currently, ce is skipped */

                if (REG_GET_FIELD(mc_umc_status, MCA_UMC_UMC0_MCUMC_STATUST0, 
UECC)
-                               == 1)
-                       amdgpu_umc_fill_error_record(err_data, err_addr,
+                               == 1) {
+                       /* loop for all possibilities of [C4 C3 C2] */
+                       for (column = 0; column < UMC_V6_7_NA_MAP_PA_NUM; 
column++) {
+                               retired_page = soc_pa | (column << 
UMC_V6_7_PA_C2_BIT);
+                               amdgpu_umc_fill_error_record(err_data, err_addr,
                                        retired_page, channel_index, umc_inst);
+
+                               /* shift R14 bit */
+                               retired_page ^= (0x1ULL << UMC_V6_7_PA_R14_BIT);
+                               amdgpu_umc_fill_error_record(err_data, err_addr,
+                                       retired_page, channel_index, umc_inst);
+                       }
+               }
        }
  }

@@ -332,8 +344,9 @@ static void umc_v6_7_query_error_address(struct amdgpu_device *adev,

                                         uint32_t umc_inst)
  {
        uint32_t mc_umc_status_addr;
-       uint64_t mc_umc_status, err_addr, retired_page, mc_umc_addrt0;
        uint32_t channel_index;
+       uint64_t mc_umc_status, mc_umc_addrt0;
+       uint64_t err_addr, soc_pa, retired_page, column;

mc_umc_status_addr =

                SOC15_REG_OFFSET(UMC, 0, regMCA_UMC_UMC0_MCUMC_STATUST0);
@@ -363,15 +376,27 @@ static void umc_v6_7_query_error_address(struct 
amdgpu_device *adev,
                err_addr = REG_GET_FIELD(err_addr, MCA_UMC_UMC0_MCUMC_ADDRT0, 
ErrorAddr);

/* translate umc channel address to soc pa, 3 parts are included */

-               retired_page = ADDR_OF_8KB_BLOCK(err_addr) |
+               soc_pa = ADDR_OF_8KB_BLOCK(err_addr) |
                                ADDR_OF_256B_BLOCK(channel_index) |
                                OFFSET_IN_256B_BLOCK(err_addr);
+               /* clear [C4 C3 C2] in soc physical address */
+               soc_pa &= ~(0x7ULL << UMC_V6_7_PA_C2_BIT);

/* we only save ue error information currently, ce is skipped */

                if (REG_GET_FIELD(mc_umc_status, MCA_UMC_UMC0_MCUMC_STATUST0, 
UECC)
-                               == 1)
-                       amdgpu_umc_fill_error_record(err_data, err_addr,
+                               == 1) {
+                       /* loop for all possibilities of [C4 C3 C2] */
+                       for (column = 0; column < UMC_V6_7_NA_MAP_PA_NUM; 
column++) {
+                               retired_page = soc_pa | (column << 
UMC_V6_7_PA_C2_BIT);
+                               amdgpu_umc_fill_error_record(err_data, err_addr,
+                                       retired_page, channel_index, umc_inst);
+
+                               /* shift R14 bit */
+                               retired_page ^= (0x1ULL << UMC_V6_7_PA_R14_BIT);
+                               amdgpu_umc_fill_error_record(err_data, err_addr,
                                        retired_page, channel_index, umc_inst);
+                       }
+               }
        }

Better to maintain the page decode logic in a single function.

umc_v6_7_save_bad_page_info(err_addr, channel)

Thanks,
Lijo

        /* clear umc status */
diff --git a/drivers/gpu/drm/amd/amdgpu/umc_v6_7.h 
b/drivers/gpu/drm/amd/amdgpu/umc_v6_7.h
index 9adebcf98582..b67677867b45 100644
--- a/drivers/gpu/drm/amd/amdgpu/umc_v6_7.h
+++ b/drivers/gpu/drm/amd/amdgpu/umc_v6_7.h
@@ -45,6 +45,10 @@
  #define UMC_V6_7_NA_MAP_PA_NUM        8
  /* R14 bit shift should be considered, double the number */
  #define UMC_V6_7_BAD_PAGE_NUM_PER_CHANNEL     (UMC_V6_7_NA_MAP_PA_NUM * 2)
+/* The C2 bit in SOC physical address */
+#define UMC_V6_7_PA_C2_BIT     17
+/* The R14 bit in SOC physical address */
+#define UMC_V6_7_PA_R14_BIT    34
  /* UMC regiser per channel offset */
  #define UMC_V6_7_PER_CHANNEL_OFFSET           0x400
  extern struct amdgpu_umc_ras umc_v6_7_ras;